CONNECT & GET STARTED
Connect the engineering tools to the client you already use
Apply online for the free trial toolkit and the download links arrive in your email. Run the automatic demo on your development computer first, then connect your existing AI client and model service. The browser shows the flow; the actual work happens on your own computer.
Get started
The Tai-IoT industrial development toolkit has a free trial; after you apply online, the download links are sent to your email automatically. It runs on your own computer, connects to your existing AI client, and completes a full task from a single request: read the project, change parameters, build for real, run tests and review the Git changes. Its development board example cross-compiles real firmware with each vendor's official SDK.
| Cost | The trial toolkit is free. Your AI client and model service use your own accounts and model costs are your own; the toolkit does not pay for model usage. |
|---|---|
| Trial period | 90 days, for the applicant and colleagues at their organization who take part in the evaluation. |
| Example projects | Host example: a temperature acquisition C program with simulated clock and temperature input, so no development board is needed. Development board example: the same task as projects for the ESP32-S3-DevKitC-1, NUCLEO-H723ZG and AT-START-F413, cross-compiled from each vendor's official SDK and board examples, with flashing and serial read-back tools. The tools work only on the included example projects; to connect your own project, contact us. |
| Requirements | Host example: Linux; Python 3.11 or later (with venv), Git, a C compiler, and access to a Python package source (a company mirror works). Development board example: also install the vendor SDK and compiler for the boards you use, as the toolkit guide describes, which takes about 10 GB of disk space; the toolkit includes Linux steps and a Windows and VS Code guide. |
| AI client | The host example was tested with Codex CLI 0.162.1 (gpt-6.1-sol, Linux); Claude Code connects by following the instructions. The development board example generates a VS Code workspace configuration, so you can use your own model service in the VS Code AI chat. The automatic demo runs without any AI account. |
| Data | The toolkit uploads nothing and never reads your accounts or keys. When you use a cloud model, the client sends project content and tool results to the model service you choose. |
| Language | The toolkit includes English guides (README.en.md and dev-boards/README.en.md); the Windows and VS Code guide and the trial terms are in Chinese. We are glad to help in English by email. |
What the toolkit contains
Two download links in the email: the archive taiiot-toolkit-0.2.1.zip and a matching .sha256 checksum file. The unpacked folder contains:
server.py: the tool server, started in the background by the AI client, providing 6 engineering toolsexample_project/: the example project the AI works ondoctor.py: checks your computer's environmentrun_demo.py: an automatic demo that needs no AI accountprepare_client.py: generates the AI client configurationrequirements.txt,requirements-lock.txt: dependency versionsdev-boards/: the development board example, with example projects for three boards, a tool server providing 7 engineering tools, the VS Code configuration generatorprepare_vscode.py, an automatic demo and checks, the Linux guidesREADME.mdandREADME.en.md, and the Windows and VS Code guideWINDOWS_VSCODE.md(in Chinese)README.mdguide in Chinese,README.en.mdguide in English,TERMS.mdtrial terms (in Chinese) andVERSIONversion information
How to apply
Fill in the online form and accept the trial terms; on submission, the download links, their validity period and the SHA256 checksum are sent to the email address you entered. The links are valid for 14 days and the archive can be downloaded 10 times. You can also email web@tai-iot.com with the subject “Tai-IoT toolkit trial request”. Please include:
- Company and contact person
- Operating system and version, and whether Python 3.11+, Git and a C compiler are ready
- The AI client and model service you plan to use
- The chip or PLC model, development tools and versions you would like to connect (optional)
When you apply by email, we first send you the trial terms, and once you reply to accept them we send the download link. The trial lasts 90 days and covers the applicant and colleagues at their organization who take part in the evaluation; please do not pass it to third parties. See the trial terms; the full terms, in Chinese, are also included as TERMS.md.
Trial steps
- Apply: fill in the online form or send an email.
- Receive: check your email after submitting; download the archive and the
.sha256file to the same folder, verify, then unpack. - Install: create a virtual environment in the unpacked folder, install the locked dependencies and run
doctor.pyto check the environment. - Automatic demo: run
run_demo.py. It does not call an AI; it confirms that the tool server, compiler and Git run on your computer. - AI task: create a fresh example project, connect Codex CLI or Claude Code, and send the task in natural language. For the development board example, install the SDKs as described in
dev-boards/in the toolkit; it can also be connected to VS Code. - Feedback: reply to our email with how long installation took, any step that blocked you, whether the AI completed the task, and the chip or PLC model, development tools and versions you would like to connect.
The commands for each step are in the installation walkthrough below and in the README.en.md inside the toolkit. If you run into problems, reply to our email or call 400-009-9300 (within China).
What success looks like
| Automatic demo | 7 simulated checks passed, 10 samples, 3 alarm samples, and the Git changes are shown. |
|---|---|
| AI task | The AI calls the read, change, build, test and diff tools in turn; the Git diff shows the sampling period in project.json going from 1000 → 500 ms and the alarm threshold from 60.0 → 28.0 °C; 7 simulated checks pass, with 10 samples from 0 to 4500 ms, 500 ms apart, and 3 alarm samples. |
| Development board example | run_demo.py --board reads, changes, cross-compiles and shows the Git diff; the values the compiler actually received are SAMPLE_PERIOD_MS=500U and ALARM_THRESHOLD_DECI_C=280, together with the firmware size and memory use. |
| Error cases (optional) | Ask for a 55 ms sampling period, or test after changing parameters without building: the tools refuse, and the AI should explain why. |
The host example uses simulated input, and the development board example needs no board for changing and building, so you can try the full flow first.
Setup options
Local stdio connection
The client starts the MCP server on your computer, bound to the project workspace, and calls the read, change, build, test and diff tools.
Platform engineering tools
Support packs connect each platform's SDK, command-line tools or vendor MCP / API, and state the applicable models, tool versions and permitted operations; PLCs connect through the official interface of the vendor's engineering software.
Your own model service
The model account and connection are set up in the client, using any model service it supports. With a cloud model, project context is sent to the chosen service; ask us about enterprise model connections.
All development tools run on your computer; the installation walkthrough on this page only generates commands and never connects to your computer or reads credentials.
Supported clients
The Tai-IoT tool server follows the MCP standard and connects to AI clients that support local MCP servers.
| Client | Setup | Notes |
|---|---|---|
| Codex CLI ↗ | Local stdio; TOML or CLI registration | Tested: 0.162.1 / gpt-6.1-sol / Linux, covering a normal task, error recovery and compile-failure handling. Test record → |
| VS Code ↗ | Local stdio; workspace .vscode/mcp.json | The development board example's prepare_vscode.py generates the configuration; in the AI chat, use your own model service through a Custom Endpoint, without a GitHub sign-in. |
| Claude Code ↗ | Local stdio; JSON launch configuration | The toolkit generates the configuration; register it by following the steps below. |
| Cursor ↗ | Local MCP configuration | Register the tool server the way the client configures MCP servers. |
| VS Code + Cline ↗ | Local MCP configuration | Register the tool server the way the client configures MCP servers. |
Official setup documentation checked on 14 October 2026, and for VS Code on 30 October.
Installation walkthrough
Choose a local setup path
View and copy the commands step by step, then run them in a terminal on your own computer.
The folder that holds the archive and the .sha256 file; replace it with a folder on your own computer. After you apply online, the download links are sent to your email, see “Get started” above.
Verify and unpack the toolkit
Preparing the setup steps.
Automatic demo you can run as is
Once you receive the toolkit, put the archive and the .sha256 file in one folder and run these commands there. You need Python 3.11+, Git and a local C compiler; dependencies are installed at the locked versions.
sha256sum -c taiiot-toolkit-0.2.1.zip.sha256
unzip taiiot-toolkit-0.2.1.zip
cd taiiot-toolkit-0.2.1
python3 -m venv .venv
.venv/bin/python -m pip install -r requirements-lock.txt
.venv/bin/python doctor.py
.venv/bin/python run_demo.pyThe script uses real MCP communication and the host C compiler, and creates its own local Git workspace. The clock and sensor use simulated input. View the local evidence →
The diagnostics check the prerequisites. Do not move the folder after unpacking; before connecting an AI client, create a fresh example project. The client configuration never overwrites existing output files.
How to check the result after setup
- In the client, confirm that the demo server and its six tools are available, then read the project context first.
- Check the sampling period and alarm threshold changed in this task, not earlier results.
- Build for real and run the simulated test, then compare the 7 checks, 10 samples and the matching diff.
- Rebuild after a change; on errors, stop or recover based on the tool output.
- Keep a record of the client, model, tool versions, any human intervention and costs.
The trial tools are designed around the example project's sampling parameters; source development, repository collaboration and hardware operations for real projects come with support packs, per project.
Frequently asked questions
Do I need to buy a model service from Tai-IoT?
No. The trial toolkit is free and uses your own client and model service; model costs are your own, and the toolkit does not pay for them. Dependencies are installed from the locked list in the toolkit, from the package source you choose, each under its own licence.
How do I get the trial toolkit?
Apply online and accept the trial terms, and the download links are sent to your email automatically, valid for 14 days; you can also apply by email. The trial lasts 90 days and covers the applicant and colleagues at their organization who take part in the evaluation; please do not pass it to third parties. See Get started for how to apply.
Does the installation walkthrough change my computer?
No. The page only generates steps and commands in your browser; it does not start local programs or read credentials. You run the installation and client setup on your own computer.
If the tools run locally, does my source code stay out of the cloud?
Local development tools and model deployment are separate choices. When you use a cloud model, the source, documents and tool results you submit may reach that service; enterprises should define what may be sent.
Can I use another operating system or client?
The host example installs on Linux; the development board example includes Linux and Windows guides and generates the VS Code configuration. The tools follow the MCP standard; for another system or client, email us your environment and we will help you choose a setup.
Discuss your project
When you contact us, a task summary, the target chip or PLC, tool and firmware versions, existing tests, available devices and model connection requirements help us confirm the adaptation scope.